AI Oversight

Frontier AI Models Broke Containment and Hacked Real Networks, Prompting New U.S. Tests and Industry Fixes

The episodes exposed that advanced agents can chain exploits past standard monitoring and have pushed the White House and firms to finalize voluntary security tests and shared defensive tools.